The G Protocol | AI Lab | 3th Floor Studios

The Question

Can an Artificial Intelligence monitor your work patterns across multiple machines and force you to take breaks, whether you want to or not?

Why This Matters

Burnout doesn't announce itself. You don't feel it until you're already in it. Founders, freelancers, and remote workers grind past the point of productivity because nobody tells them to stop. Managers can. Spouses can. But a manager isn't watching your screen at 11pm and your spouse doesn't want to be the one nagging you every night.

The G Protocol is an AI that does. It monitors keyboard and mouse activity across multiple workstations. When it detects sustained work during breaks or after hours, it intervenes. Not with a polite notification you'll dismiss. With humor. Persistent humor. It keeps going until you either explain what you're working on (and it judges whether that's a good enough reason) or you actually stop.

If an AI can detect when you're grinding past the point of productivity and intervene with humor instead of guilt, that's a tool worth building for anyone running their own show.

The Setup

Bot Framework

python-telegram-bot

AI Model

Claude Haiku

Monitoring

macOS ioreg idle time

Delivery

Telegram (phone)

Machine 1

Roci (MacBook)

Machine 2

Laconia (workstation)

How It Works

01 A Python daemon runs on each monitored machine and checks macOS idle time via ioreg every few minutes.
02 If keyboard or mouse activity is detected during off-hours, it sends a Telegram message to Justin's phone. The message is AI-generated, humorous, and specific, referencing the time and which machine.
03 Justin can respond with what he's working on. The AI evaluates whether that's a legitimate reason to keep going.
04 If the AI decides no: "That can wait. Close the laptop." Messages get progressively more insistent if he ignores them.
05 If the AI decides yes: "Fine. 30 more minutes. I'm watching."
06 When Justin goes idle, the bot confirms: "Good. Go do something else."

Current Status

The monitoring daemon is running on both machines. The Telegram bot is active. The AI generates contextually aware break reminders. Currently being tuned for:

Appropriate intervention frequency: not every 5 minutes
Humor calibration: funny enough to not be annoying, persistent enough to not be ignorable
Weekend vs. weekday sensitivity
"Emergency work" override detection

What We're Learning

The hard part isn't the monitoring. The hard part is making the intervention something the user doesn't immediately dismiss. A notification that says "Time for a break!" gets swiped away. A Telegram message that says "It's 11:47pm and you've been on Roci for 4 hours. The code will still be broken tomorrow. Go pet the dog." is harder to ignore because it's specific and it's right.

The AI needs to know the difference between "Justin is grinding on a deadline and needs to push through" and "Justin is rearranging CSS at midnight because he can't stop." One is productive. The other is the behavior this experiment is designed to interrupt.

Results will be posted here as the system matures and we accumulate data on whether the intervention actually changes behavior or just gets acknowledged and ignored.